This is my personal question about a general statistical matter. This is not a question about JMP from users. It might not be appropriate to ask this kind of question here, but I cannot find the other better place where I should ask inside JMP group. The question is strongly related to my post two years ago (Simple and honest explanation for two-sample t test? - JMP User Community).
Are the flowchart statistical analyses still survived in your country?
Around 60's-90's, we saw some sacred and inviolable ritual rules for performing statistical testing in many cookbooks. (I think many of them are written by non-statisticians like medical researchers, sociologists, psychologists and so on.) For example, the ritual for comparing two independent samples was like the below.
(1) Perform normality test for each group.
(2) If one of two tests is significant, perform Wilcoxon's test.
(3) If not, perform homogeneous test for variances.
(4) If it is significant, perform Welch's test.
(5) If not, perform Student's t test.
These flowchart rituals crihave beenticized strongly by many statisticians.
The critic for all general flowchart procedures is written on, for example, pp.1-4 in McElreath, R. (2020:2nd ed.) Statistical Rethinking: A Bayesian Course with Examples in R and Stan (2nd ed.),CRC Press.
Speaking to Welch, Student and Wilcoxon test, please try the following prompt in Copilot, ChartGPT and so on. There are too many papers, so I cannot select one of them.
----------------------
Could you pick up some papers to criticize the flowchart procedure, to perform Wilcoxon test if normality test is significant, to perform Welch test if homogeneous test for variances is significant, to perform Student's test if not?
And, could you also pick up some papers to explain that Student's t test is approximately equivalent to a permutation test, and can be used as a nonparametric test for testing H0:F1(y) = F2(y) for all y (although as a nonparametric test, Student's t test does not become a test for H0:mu1=mu2).
----------------------
As regard to preliminary normality test before t-tests, John Sall-san wrote about the robustness of t-test about 20 years ago in the JMPer Cable’s article with a unique title.
Leptokurtosiphobia: Irrational Fear of Non-Normali
As regard to preliminary homogenous test for variances before t-tests, G.E.P Box’s famous word, "To make the preliminary test on variances is rather like putting to sea in a rowing boat to find out whether conditions are sufficiently calm for an ocean liner to leave port!", is quoted in SAS’ GLM Procedure help (SAS Help Center: Homogeneity of Variance in One-Way Models).
As regard to analyzing DOE data, G.E.P. Box also says “”Block what you can and randomize what you can’t” can approximately justify an anslysis “as if” standard assumptions were true”.
In Japan, I think the flowchart statistical analysis is still copied and pasted by many non-statisticians, bloggers or "influencers" in their internet telephone game. Although I am afraid my perspective is biased, I think only a very few Japanese statisticians recommend the flowchart ritual anymore.
I watched the presentation by David Meintrup-san at Discovery Summit Europe 2026. I am very surprised the above flow-chart statistical analysis is shown as the expert's statistical analysis. I am also afraid that LearnBot and Assistant "learned" this kind of rituals from net memes in blog articles written by non-statisticians.
